Appraisal Navigator
Navendu Garg
Illinois Institute of Technology 10 W 31st State Street Chicago, Illinois 60616

Kenneth Bloom
Illinois Institute of Technology 10 W 31st State Street Chicago, Illinois 60616

Shlomo Argamon
Illinois Institute of Technology 10 W 31st State Street Chicago, Illinois 60616

gargnav@iit.edu

kbloom1@iit.edu

argamon@iit.edu

ABSTRACT
Much interesting text on the web consists largely of opinionated or evaluative text, as opp osed to directly informative text. The new field of `sentiment analysis' seeks to characterize such asp ects of natural language text, as opp osed to just the bare facts. We suggest that `appraisal expression extraction' should b e viewed as a fundamental task for sentiment analysis. We define an `appraisal expression' to b e a piece of text expressing some evaluative stance towards a particular ob ject. The task is to find these elements and characterize the typ e and orientation (p ositive or negative) of the evaluative stance, as well as its target and p ossibly its source. Potential applications of these methods include new approaches to the now-traditional tasks of sentiment classification and opinion mining, as well as p ossibly for adversarial textual analysis and intention detection for intelligence applications. Categories and Sub ject Descriptors: H.3.3 Information Storage and Retrieval:Information Search and Retrieval, I.2.7 Artificial Intelligence:Natural Language Processing [Text analysis], H.3.1 Information Storage and Retrieval:Content Analysis and Indexing [Linguistic processing] General Terms: Algorithms, Exp erimentation Keywords: Appraisal Extraction, Opinion Mining, Sentiment Analysis, Shallow Parsing, Appraisal Theory

Once the system extracts the attitude groups and the appraised things, it tries to associate each attitude group with a target using hand-selected list of acceptable paths through the deep dep endency parse tree of a sentence. -- -- -- appraised - -  x  - y  - appraisal This linkage selects the sub ject of a sentence like where the appraisal modifies a noun in the predicate (e.g. "The Matrix" in "The Matrix is a very good movie."). Each extracted expression is analyzed to give a high-level generic representation of the meaning of the appraisal expression in terms of its evaluative function in the text. For example, in "I found the movie quite monotonous" the sp eaker (the Source) adopts a negative Attitude (`quite monotonous') towards the Target (`the movie'). Using data mining we extracted association rules to determine which asp ects of a target p eople like or dislike, and in which ways.
nsubj dobj amod

2. APPRAISAL NAVIGATOR
Appraisal Navigator1 is a web interface for the prototyp e system. Using the interface a user can submit three typ es of queries: 1. Query to find relevant appraisal expression rules. 2. Query to find relevant documents for a given appraisal expression and relevant keywords. 3. Query to find relevant sentences for a given appraisal expression and relevant keywords. A typical query, comprising an appraisal expression like (attitude = affect, orientation = p ositive, appraised = digital cameras) and relevant keywords like `excellent' or `amazing', will return the matched rules. The user can hover on each rule and see the related documents or sentences. Further, the user can select one of the rules and re-submit it as a new query. This user-interface can function as useful tool for detailed analysis of public opinion ab out various targets.

1.

INTRODUCTION

The task of `sentiment analysis' is to find and characterize asp ects of opinionated natural language. We define an `appraisal expression' to b e a piece of text expressing some evaluative stance towards a particular target. By analogy to information extraction, we consider representing appraisal expressions as frames filled with several values for several slots. Thus, an appraisal expression comprises: Source, Attitude, and Target. We have develop ed a prototyp e system for navigating in a b ody of texts based on extracted appraisal expressions. Currently, the system extracts adjectival appraisal expressions only. To extract an appraisal expression, the system first finds adjectival appraisal groups[1](a head adjective with defined attitude type and an optional preceding list of appraisal modifiers ). It then extracts the target groups by matching phrases in the lexicon to the phrases in the text and assigns the target typ e assigned in the lexicon.
Copyright is held by the author/owner(s). SIGIR'06, August 6­11, 2006, Seattle, Washington, USA. ACM 1-59593-369-7/06/0008.

3. REFERENCES
[1] Casey Whitelaw, Navendu Garg, and Shlomo Argamon. Using appraisal groups for sentiment classification. In Proc. Conference on Information and Know ledge Management, Bremen, Germany, 2005.

1

http://lingcog.iit.edu/appraisalnavigator/

727